I bought some white PVC piping the other day and started fiddling and testing, making various perspex diffusers of different opaqities (positioned just before the first bend), and entrance hole reducers with aluminium tape on the inside (both do help significantly, in making light bounce around a bit in the entrance cavity) , and I must say that it is a real challenge to make this system really integrate. And everything after the first bend does not add to integration, extra bends mostly reduce the light. In the end I got reasonably recognisable measurements for the common reflector flashlights (if pointed straight down on the diffusor!), but a zoomie in flood modus was over 25% off. Although it provides ballpark numbers, I’d say it is more of a light reducer than an integrator. Unexpectedly, my pipe design does not change CCT that much but does increase duv enormously. If you look at the spectrum I can see why: both the blue peak is reduced, and a broad band around 600nm, result: green. The portion over 630 nm is increased again. This may all be the result of the specific type of PVC that I got (it looked plain white though). I may do a post about my findings, but then, maybe not. My conclusion: I stick to spheres for integration 
Cool info, djozz! I’ve also wanted to make a PVC tube for checking how they integrate.
That is not such a great design. Firstly, don’t use a black cloth, that absorbs a lot of light. You want to use something reflective to bounce the light towards the sensor.
Second is your not really blending up or diffusing the light. A tight focused beam light is probably going to read higher than a wide focused beam. You may be getting direct reflections on the sensor which will make the readings wonky. The middle section doesn’t need to be so long. Just having more bends will help break up the light beams. Try removing the end where the sensor is and shine your light in normally. See if the area where the sensor is gets “smooth” light or splotchy light. You’ve seen the Texas_Ace Lumen Tube, right? Link is in my sig. It’s got more bends plus 3 translucent diffusers to help smooth out the light before it gets to the sensor. This smoothing or “integrating” of the light is the key job of an integrating sphere. Yours is severely lacking in this department.
Lumen tubes are also sensitive to distance. Having a piece of glass that you sit the light is convenient, but it also moves the led closer or further away depending on the reflector or optic depth. This leads to inconsistencies. This doesn’t matter as much with ball shaped integrating spheres. Try making those other changes and see if using the glass is okay for you or not.
I just got the three SST20 4000K CRI90 Convoy S2. So two more BLF348 S2+ kits are ready to go after one S2+ was reserved for M4D M4X.
I recenltly made one myself and indeed, getting it accurate for all type of lights is VERY difficult.
However, i ended up with a setup which is quite satisfying and quite accurate
And yes, welcome to the world of real lumens. Only just now i realize how unreliable the human eye seems to be.
I did not realize how fast output drops, how much effect temperature, cell type, condition and charging percentage affects the output. Now i do.
It really is. I fiddled with mine for quite a while trying to get the results to be more consistent and less dependent on factors like beam shape. The end result could certainly still use some improvements, but it’s at least okay-ish now.
What I did to make it work better was giving it something similar to one-way valves. Using some biased diffuser sheets which allow more light to pass in one direction than in the other seemed pretty helpful. There is one right at the entry, pointing in. Another just after the first bend, pointing forward. Then one a few cm before the sensor, pointing backward. Aside from diffusing the light three times, this also seems to help reduce the throw bias and get the light to bounce around more before it hits the sensor.
For calibration, I’m using a maukka S2+, a maukka BLF-348, and a ROT66-219 from the same batch as maukka’s. The correction factor is set so that all three lights measure within 2% of expected values. But I expect my actual margin of error is much, much larger… mostly because it’s not a sphere.
Comparing tube results to my old milk carton, the tube gets values anywhere from 75% to 110% of what the milk carton says… with the highest numbers being throwy lights and the lowest numbers being floody lights. An Emisar D1S, for example, measures 110% between the two, while a ROT66 measures 85%.
So the tube certainly seems to favor throwy lights. Or the milk carton favors floody lights. Or both. Probably both.
That’s probably the right conclusion. 
Thanks for the advice! I’m going to head to the hardware store tomorrow and pick up some new, curvier PVC tubing and have a go at building a new one. I’ve also ordered some light diffuser sheets from eBay which I’ll add along the light path too.
How would you design the light “input” end? I’ve seen some people use foam as well.
At least I have calibrated lights now! Watch this space…
It’s definitely tricky. You want an opening sized just right. Both my TA Tube and JoshK Sphere have large openings with 4 or 5 adapters that have different inner diameter holes. This seems to work well. It also allows you to move the flashlight head in and out of the opening.
I’d like to have a big 6” adjustable aperture that was chrome plated on one side.
I don’t think they are made that big, though. Probably really expensive if they were available.
The latest revision of the Bobbersphere. Still not sure how well it does on different beam types, waiting on a more throwy light from maukka.
Extech LT45 meter with custom factor setting and laser cut discs that Texas_Ace uses on his tubes:
Internal baffle + diffuser:
OK so a quick redesign of my tube to make it bendier, and I have the following multiplication factors. There’s two sheets of diffuser in the middle.
BLF 348: 0.113
S2+: 0.126
With these multipliers, it gives the following results which are all too low. All are full power / turbo on freshly charged high current batteries. I’m pointing the light just inside the entrance and using the white cutout to close the hole a bit.
Olight M2R Neutral White: 1379lm
Olight X7R Marauder: 8316lm
Emisar D4S XP-L HD 3D: 3480lm
Emisar D4V2 SST-20 5000k: 3108lm
Improvement suggestions welcome.
It seems that your setup is biased towards cool (or low CRI) or throwy since the factor is much higher on the S2+. Also this would explains why the wide floody beam of the optics in the Emisars read low.
Did you use the average of those two correction factors for calculation. The S2+ factor would be more pertinent to the Olight M2R calculation. That one doesn’t seem too far off.
I changed it to the average now and it still shows the M2R as 1458 which is about 150lm low, and the X7R as 7887 which is about 4000lm low!
With the S2+ as the calibration reference, the M2R would read 1537 lm which would be pretty much bang on. I measured it at 1500 lumens which is what Olight also specifies it at. The X7R is much floodier, which explains part of the error, but my sample wasn’t near 12000 lumens either.
Thanks. So you think my design is biased too much towards lights with a tighter hotspot?
Measuring my BLF A6 yields 1172lm
My Astrolux C8 showed 1373lm
My S2+ with Samsund LH351D only 801lm when it should be a lot higher. But that’s neutral white and a lot floodier
It isn’t as consistent as I’d like 
My S2+ with 8x3175 and XP-L HI does 780 lumens so 800 lumens for LH351D seems reasonable. My A6 with XP-L HD 3D is 1200 lumens.
My LH351D S2+ has an A6 driver… Should have said!
My A6 has a cool white XP-L HI
Wow, thanks for sharing this
My recently built lumen tube was reading almost the same numbers with a S2+ XPL HI-4B with 8*7135
I thought it was reading a bit low, but for other lights with given numbers, all my measurements weren’t to far off. It seems my setup is better than i thought !
I have tweaked the setup, this was the initial build
I’m pretty sure his LX1330B lux meter is a bit more sensitive to cool white than it is to warm white (same for HS1010A). If your testing a cool white light, use the 0.126 factor and if measuring a warm white use the 0.113 factor. If the light is neutral white, use a factor like 0.120.
Most of my lights are CW with a few NW, so I set my lumen tube for CW readings. If I do measure a NW light I try to remember to add a little more to compensate. I forget the correct number, but something like 3 percent. WW might read 5-6 percent low.
That certainly looks better. You say you put two sheets of diffuser material in the middle? They are probably acting like a single sheet. It would probably be better to spread them out. That way you have partially diffused light coming from the first one and then hitting the 2nd sheet to further diffuse it.
I am trying to find out what others have measured the x7r turbo at. 31% below spec seems pretty big, but might be accurate. Manufacturers often over rate their products output, but I thought olight would be a bit closer than this.
Okay, it seems Maukka has tested the x7r. It was tricky for him to measure because the proximity sensor would reduce output. He measured 3 lower levels in his sphere, then measured all levels outside the sphere, I’m guessing like a ceiling bounce. Then he matched up the 3 lower levels and calculated the upper levels. He also says the 12,000 lumen rating from Olight is at Turn On, not 30 seconds. :disappointed:
Maukkas Turbo S measurements at 30 seconds were 9,700 lumen. Your still reading about 14% lower than him. Maybe the proximity sensor is reducing output? IDK.
M4DM4X shows his honking tube which seems to be reading in line with the calibration lights. Good job!